Registration techniques (both rigid and non-rigid) are employed in many scene perception systems, including robotics, navigation, and virtual reality applications, where objects in the scene can be semantically understood by associating them with known object models. Registration of two 3-dimensional objects requires an estimation of the translation, rotation, and scaling between the objects, such that an alignment can be determined between the scanned object and the object model. Some existing techniques perform a coarse registration between objects, but are sensitive to large differences in the initial orientations of the objects relative to each other. Some such techniques assume that there is no significant in-plane rotation and will fail if that is not the case. Other techniques rely on assumed geometric constraints, and can be highly sensitive to noise in the scan measurements.
Although the following Detailed Description will proceed with reference being made to illustrative embodiments, many alternatives, modifications, and variations thereof will be apparent in light of this disclosure.